Skip to content

The fine print

How we use AI

The judge is an AI, and a very solemn joke. Here is every place we use AI, what it sees, and where a person makes the call.

Last updated November 15, 2026

  1. Judge Pettybone is an AI

    Verdicts, sentences and appeal rulings in petty cases are written by an AI playing Judge Pettybone. They are entertainment. They are not legal advice, have no legal effect, and shouldn’t be used to make real decisions. The public jury’s votes are real people; the judge isn’t.

    Della Docket’s recaps, the Case of the Day rulings, the closing line on a poll or a question, and similar commentary are AI-written too, from what is already public on the case or post. The judge and Della are instructed never to declare a real, identifiable person guilty of a real-world crime.

  2. Who runs the AI

    Our AI features use large language models provided through our hosting platform’s AI service. They process requests on our instructions to produce each feature’s output. Gavel Drop doesn’t train AI models, and we don’t use your content to build an advertising profile.

  3. What the AI sees

    • A verdict: the case’s testimony, witness statements, objections and exhibit captions. Exhibit photos themselves are not sent.
    • Moderation: the reported post and a little context around it (for example the title and two or three comments before it), with ages and public-figure status as yes / no facts. Photos in direct messages are never sent to the moderation reviewer.
    • Photo checks: before photos appear in comments, private groups, messages and ads, an AI screen checks them for content our rules don’t allow. Photos are stripped of hidden details such as location first.
    • Message safety: some direct messages (for example from new accounts, with links, or in a conversation with a member under 18) may be checked by an AI for scams and threats, with only the message text and no names, ids or ages.
    • Things you ask for: when you tap a helper (translate, “Sharpen my debate”, Ask the Clerk), only what that helper needs is sent.
  4. The AI moderation reviewer

    Reports first go to an AI reviewer that checks them against our posted rules. It keeps posts that are clearly lawful, and removes only a narrow set of clearly prohibited content when it is very confident. Hard calls go to a person: anything involving a child’s safety, intimate images, threats, copyright, legal demands, removing a public figure’s or a member under 18’s post, or anything the AI is unsure about.

    If the AI removes your post, you can appeal: a second, independent review looks again, and if it disagrees the decision goes to a person. A sample of the AI’s decisions is also checked by a person, and a rule is paused if the AI keeps getting it wrong. Reports of child sexual abuse material skip the AI entirely and go straight to a person.

  5. Ask the Clerk

    Our help assistant answers only from our help articles, and every answer points to the articles it used. If it can’t find an answer, or the question is about your account, a payment dispute, a legal matter or someone’s safety, it hands you to a person instead. It can’t take actions on your account. We keep a short record of the outcome (for example which articles it pointed to), not your question.

  6. Translation and drafting helpers

    • Translate turns a post or comment’s public text into your language. Translations are saved and shared with other readers of the same text.
    • Drafting helpers suggest clearer wording for a debate question and its two sides, a Jury Note, or an advertiser’s ad copy. A suggestion is used only if you choose it, and it is screened like anything else you post.
    • Settle it can turn a chat you paste into a draft case. Contact details are removed before the AI sees it (if you share screenshots, the AI reads the text in them), the chat itself isn’t stored, and the draft is deleted when you file or discard it, or after 24 hours.
    • Our own team uses AI drafts for example cases, I Got Questions producer notes and marketing posts. A person edits and approves them before anything is published.
  7. Limits

    AI makes mistakes. A ruling may be unfair, a translation may be off, and a summary may miss something. Report anything that looks wrong, and contact us if an AI decision about your content or account seems mistaken. Teams sessions don’t use AI: their rulings come from fixed templates.